It is frequently difficult to identify social learning in natural animal populations, or even in captive groups of animals. We are developing new mathematical and statistical methods to do this.

## The Option Bias Method

The option-bias method is based on the well-established premise that social learning will generate greater homogeneity in behaviour between animals than expected in its absence. The procedure compares the observed level of homogeneity to a sampling distribution generated utilizing randomization and other procedures, allowing claims of social learning to be evaluated according to consensual standards. We have established that the method has higher statistical power than conventional inferential statistics, which is likely to be important in cases where sample size is small. We envisage that the method could be of value in both in assessing the validity of claims for culturally transmitted behaviour, and in enabling investigation of the social learning strategies deployed. For further details see:

To implement these analyses, you need to:

- download the statistical package R, available free from http://www.r-project.org/
- download Option Bias files “OB example” and “OB functions” in zip archive, and unzip
- cut and paste the code contained in the file “OB functions” into the R console.

The file “OB example” provides an example Option Bias analysis which can be used as a guide to using the functions.

## The Boogert et al. (2008) Randomisation Method

It is often assumed that social transmission tends to occur more rapidly between closely associated individuals in a population. Where this is the case we can test for social transmission by testing whether the order in which individuals acquire a behavioural trait is related to a matrix of measured associations between them. One way in which this can be done is by calculating an appropriate test statistic, and then generating a null distribution by randomising the order of acquisition a large number of times, calculating the test statistic each time. For further details see:

To implement these analyses, you need to:

- download the statistical package R, available free from http://www.r-project.org/
- download Boogert Randomisation files zip

archive and unzip - cut and paste the code contained in the file “Boogert Randomisation” into the R console

The file “Boogert Randomisation example” provides an example analysis that can be used as a guide to using the functions.

## Network Based Diffusion Analysis (NBDA) Version 1.2

NBDA, first invented by Franz & Nunn (2009), infers social transmission when the pattern of acquisition of a novel behavioural trait follows the network of associations between individuals. In this respect, the method is similar to Boogert et al’s randomisation method, but has a number of advantages that are discussed in Hoppitt et al (2010). Franz & Nunn’s original method requires data on the times at which individuals acquire a novel behavioural trait. In our paper (Hoppitt et al, 2011) we expand their method and introduce our own variant which requires data only on the order in which individuals acquire the trait. We call these methods Time of Acquisition Diffusion Analysis (TADA) and Order of Acquisition Diffusion Analysis (OADA) respectively. The relative advantages and disadvantages of each method are discussed in Hoppitt et al (2010).

Here we provide R code to implement both OADA and our extended version of TADA.

To implement these analyses, you need to:

1. download the statistical package R, available free from http://www.r-project.org/

2. download NBDA files in zip archive and unzip:

- NBDA code 1.2.13.R: R code
- NBDA User Guide V1.2.1.pdf: instructions on how to use the code
- starlingSolve1.2 and stratDiffs: example dataset
- Example use of profLik functons.R

3. cut and paste the code contained in the file “NBDA code 1.2.13.R” into the R console

## Bayesian NBDA

We have expanded NBDA to allow it to quantify the evidence for social transmission across a number of diffusions. To do this it is important to account for individual differences in asocial learning by fitting models with appropriate random effects. As random effects can be difficult to implement using maximum likelihood methods, here we take a Bayesian approach. See Nightingale et al (2014) for technical details.